Reka Flash 3

Reka · reka/reka-flash-3

Provider	Reka
Family	reka
Type	llm-chat
Status	active
Released	2025-03-11
Updated	2025-03-13
Parameters	20.9B
Open weights	yes

What it fits on

Ordered by predicted speed. Bandwidth sets decode rate; memory decides whether it runs at all.

Every figure here is computed, not measured. Fit is weights at each quantisation against device memory, with a 25% allowance for the KV cache, activations and the OS. Decode rate is the memory-bandwidth roofline at 70% efficiency. Nobody has run this model on these devices.

Device	Memory	Bandwidth	Best quality	Weights	tok/s	Fastest
Cerebras WSE-3	44.0 GB	21000000.0 GB/s	fp8	20.91 GB	~703164.8	~1406329.7 q4
NVIDIA GB200 Grace Blackwell Superchip	372.0 GB	16000.0 GB/s	bf16	41.81 GB	~267.9	~1071.5 q4
AMD Instinct MI355X	288.0 GB	8000.0 GB/s	bf16	41.81 GB	~133.9	~535.7 q4
NVIDIA B200	180.0 GB	7700.0 GB/s	bf16	41.81 GB	~128.9	~515.7 q4
Google TPU7x (Ironwood)	192.0 GB	7380.0 GB/s	bf16	41.81 GB	~123.6	~494.2 q4
AMD Instinct MI325X	256.0 GB	6000.0 GB/s	bf16	41.81 GB	~100.5	~401.8 q4
AMD Instinct MI300X	192.0 GB	5300.0 GB/s	bf16	41.81 GB	~88.7	~354.9 q4
NVIDIA H200 SXM	141.0 GB	4800.0 GB/s	bf16	41.81 GB	~80.4	~321.4 q4
NVIDIA H100 NVL	94.0 GB	3900.0 GB/s	bf16	41.81 GB	~65.3	~261.2 q4
NVIDIA H100 SXM	80.0 GB	3350.0 GB/s	bf16	41.81 GB	~56.1	~224.3 q4
AMD Instinct MI250X	128.0 GB	3200.0 GB/s	bf16	41.81 GB	~53.6	~214.3 q4
Google TPU v5p	95.0 GB	2765.0 GB/s	bf16	41.81 GB	~46.3	~185.2 q4
NVIDIA A100 80GB SXM	80.0 GB	2039.0 GB/s	bf16	41.81 GB	~34.1	~136.5 q4
NVIDIA H100 PCIe	80.0 GB	2000.0 GB/s	bf16	41.81 GB	~33.5	~133.9 q4
NVIDIA GeForce RTX 5090	32.0 GB	1792.0 GB/s	fp8	20.91 GB	~60.0	~120.0 q4
Google TPU v6e (Trillium)	32.0 GB	1638.0 GB/s	fp8	20.91 GB	~54.8	~109.7 q4
AMD Instinct MI210	64.0 GB	1600.0 GB/s	bf16	41.81 GB	~26.8	~107.1 q4
NVIDIA A100 40GB SXM	40.0 GB	1555.0 GB/s	fp8	20.91 GB	~52.1	~104.1 q4
Google TPU v4	32.0 GB	1200.0 GB/s	fp8	20.91 GB	~40.2	~80.4 q4
NVIDIA GeForce RTX 4090	24.0 GB	1008.0 GB/s	q6	15.68 GB	~45.0	~67.5 q4
AMD Radeon RX 7900 XTX	24.0 GB	960.0 GB/s	q6	15.68 GB	~42.9	~64.3 q4
NVIDIA GeForce RTX 5080	16.0 GB	960.0 GB/s	q4	10.45 GB	~64.3	~64.3 q4
NVIDIA GeForce RTX 3090	24.0 GB	936.0 GB/s	q6	15.68 GB	~41.8	~62.7 q4
NVIDIA L40S	48.0 GB	864.0 GB/s	fp8	20.91 GB	~28.9	~57.9 q4
Apple M3 Ultra	512.0 GB	819.0 GB/s	bf16	41.81 GB	~13.7	~54.8 q4
AMD Radeon RX 7900 XT	20.0 GB	800.0 GB/s	q5	13.07 GB	~42.9	~53.6 q4
Apple M2 Ultra	192.0 GB	800.0 GB/s	bf16	41.81 GB	~13.4	~53.6 q4
Google TPU v5e	16.0 GB	800.0 GB/s	q4	10.45 GB	~53.6	~53.6 q4
NVIDIA GeForce RTX 4080 SUPER	16.0 GB	736.0 GB/s	q4	10.45 GB	~49.3	~49.3 q4
NVIDIA GeForce RTX 4070 Ti SUPER	16.0 GB	672.0 GB/s	q4	10.45 GB	~45.0	~45.0 q4
AMD Radeon RX 9070 XT	16.0 GB	640.0 GB/s	q4	10.45 GB	~42.9	~42.9 q4
Apple M4 Max	128.0 GB	546.0 GB/s	bf16	41.81 GB	~9.1	~36.6 q4
Apple M1 Max	64.0 GB	400.0 GB/s	bf16	41.81 GB	~6.7	~26.8 q4
Apple M2 Max	96.0 GB	400.0 GB/s	bf16	41.81 GB	~6.7	~26.8 q4
Apple M3 Max	128.0 GB	400.0 GB/s	bf16	41.81 GB	~6.7	~26.8 q4
NVIDIA L4	24.0 GB	300.0 GB/s	q6	15.68 GB	~13.4	~20.1 q4
NVIDIA GeForce RTX 4060 Ti 16GB	16.0 GB	288.0 GB/s	q4	10.45 GB	~19.3	~19.3 q4
Apple M4 Pro	64.0 GB	273.0 GB/s	bf16	41.81 GB	~4.6	~18.3 q4
Apple M4	32.0 GB	120.0 GB/s	fp8	20.91 GB	~4.0	~8.0 q4

Where it runs

Taken from the card's availability section.

Platform	Id on that platform
Hugging Face	RekaAI/reka-flash-3

Verified benchmark evidence

Each row was checked against its source by a reviewer, and carries the model identifier as actually evaluated — which is not always the same as this card's.

Benchmark	Evaluated as	Score	Evidence date	Source kind
aa_lcr	Reka Flash 3	0.0%	2026-09-10 evaluated	independent evaluator	source
critpt	Reka Flash 3	0.0%	2026-09-10 evaluated	independent evaluator	source
gpqa_diamond	Reka Flash 3	52.93%	2026-09-10 evaluated	independent evaluator	source

Reported benchmark scores

This card reports no benchmark scores yet.

Benchmark	Catalogue standing	Score

Data

This card as JSON · See it in the graph · Edit on GitHub